Early Detection At-Risk Students using Machine Learning

Machine Learning is one of the most popular technologies using in many industries, especially to analyze the data and find key insight or new knowledge. In education industry, many studies have applied machine learning techniques for various purposes. One important area is to early detect at-risk students by using data from various sources such as log data from learning management systems (LMSs), class attendances, and actual score from both formative and summative assessments. We present a comparative study aiming to find the most important features and the best classification algorithms to classify at-risk students based on they behaviors. The data are collected from Moodle system [1], printing services system, and students grad system at one of the faculty in the university. The experiment results are evaluated in terms of overall accuracy, precision, and recall. The random forest with oversampling on minority class shows the best result. The performances of the models is better when we have more data in each week of the semester. During week 5, the model can detect about 74 percent of at-risk students.